Gemma 3 12B
Gemma 3 12B needs 8.7 GB of VRAM at Q4_K_M with 8k context, computed from its config.json as checked on 16 October 2026. The smallest card that holds it has 12 GB, and the fastest desktop card that runs it, the RTX 5090, generates at most 177 tokens per second.
- Parameters
- 12.2B
- Context
- 131,072 tokens
- Licence
- Gemma Terms of Use
- Released
- March 2025
- Config checked
- 16 October 2026
Open Gemma 3 12B in the LLM checker
Which card size runs it
8.7 GB against each memory size. A model may use 95% of a card, and only what is left over, or missing, is coloured; tight means above 85%, with little room for a longer conversation.
Other quantisations, context lengths or a Q8 KV cache: the VRAM calculator.
Computed 16 October 2026Which GPU to pick
The smallest card here that loads it at Q4_K_M with 8k of context is the RTX 3060 (12 GB): 8.7 GB needed, an estimated 36 tokens per second. For room to work, meaning 32,768 tokens of context while staying under 85% of memory, step up to the RTX 4060 Ti (16 GB), which needs 10.3 GB for that and generates at about 28 tokens per second.
At Q8_0, which is near-lossless, the card is the RTX 4060 Ti (16 GB), needing 14.8 GB. On Apple silicon, a Mac mini M4 Pro holds it with 32,768 tokens of context at an estimated 27 tokens per second; prompt processing is slower than on Nvidia, which long documents make obvious. Among unified-memory desktops, the Ryzen AI Max+ 395 holds it at 32,768 tokens in 10.3 GB, at an estimated 25 tokens per second.
Speed on each card
| GPU | Fits | Est. tokens/s | Feels like | Memory | Bandwidth |
|---|---|---|---|---|---|
| H100 SXM | Yes | ≤ 331 | faster than you can read | 80 GB | 3,350 GB/s |
| A100 | Yes | ≤ 202 | faster than you can read | 80 GB | 2,039 GB/s |
| RTX 5090 | Yes | ≤ 177 | faster than you can read | 32 GB | 1,792 GB/s |
| RTX PRO 6000 Blackwell | Yes | ≤ 177 | faster than you can read | 96 GB | 1,792 GB/s |
| RTX PRO 5000 Blackwell | Yes | ≤ 133 | faster than you can read | 48 GB | 1,344 GB/s |
| RTX 4090 | Yes | 100 | faster than you can read | 24 GB | 1,008 GB/s |
| RTX 5080 | Yes | 95 | faster than you can read | 16 GB | 960 GB/s |
| RX 7900 XTX | Yes | 95 | faster than you can read | 24 GB | 960 GB/s |
| RTX 6000 Ada | Yes | 95 | faster than you can read | 48 GB | 960 GB/s |
| RTX 3090 | Yes | 93 | faster than you can read | 24 GB | 936 GB/s |
| RTX 5070 Ti | Yes | 89 | faster than you can read | 16 GB | 896 GB/s |
| RTX PRO 4500 Blackwell | Yes | 89 | faster than you can read | 32 GB | 896 GB/s |
| L40S | Yes | 85 | faster than you can read | 48 GB | 864 GB/s |
| RX 7900 XT | Yes | 79 | faster than you can read | 20 GB | 800 GB/s |
| RTX A6000 | Yes | 76 | faster than you can read | 48 GB | 768 GB/s |
| RTX 4080 SUPER | Yes | 73 | faster than you can read | 16 GB | 736 GB/s |
| RTX 5070 | Yes | 66 | faster than you can read | 12 GB | 672 GB/s |
| RTX PRO 4000 Blackwell | Yes | 66 | faster than you can read | 24 GB | 672 GB/s |
| RX 9070 XT | Yes | 63 | faster than you can read | 16 GB | 640 GB/s |
| Radeon AI PRO R9700 | Yes | 63 | faster than you can read | 32 GB | 640 GB/s |
| Arc Pro B70 | Yes | 60 | faster than you can read | 32 GB | 608 GB/s |
| RTX 4070 | Yes | 50 | comfortable for chat | 12 GB | 504 GB/s |
| Arc B580 | Yes | 45 | comfortable for chat | 12 GB | 456 GB/s |
| Arc Pro B60 | Yes | 45 | comfortable for chat | 24 GB | 456 GB/s |
| RTX 5060 Ti | Yes | 44 | comfortable for chat | 16 GB | 448 GB/s |
| RTX PRO 4000 Blackwell SFF | Yes | 43 | comfortable for chat | 24 GB | 432 GB/s |
| RTX 3060 | Yes | 36 | comfortable for chat | 12 GB | 360 GB/s |
| RX 9060 XT | Yes | 32 | comfortable for chat | 16 GB | 320 GB/s |
| RTX Spark | Yes | 30 | comfortable for chat | 128 GB (102 for the GPU) | 300 GB/s |
| RTX 4060 Ti | Yes | 28 | comfortable for chat | 16 GB | 288 GB/s |
| DGX Spark | Yes | 27 | comfortable for chat | 128 GB (126 for the GPU) | 273 GB/s |
| Ryzen AI Max+ 395 | Yes | 25 | comfortable for chat | 128 GB (96 for the GPU) | 256 GB/s |
Single-stream decode, from memory bandwidth at 70% efficiency. Prompt processing and batching not included. These are ceilings, not measurements: Gemma 3 12B reads only 7.1 GB of weights per token at Q4, so little that the per-token costs the formula leaves out decide much of the real speed. The 5 figures marked ≤ are where a real runtime falls furthest below the number shown. See the speed estimator for other quantisations and Apple hardware, or every card compared if you are choosing hardware rather than a model.
Computed 16 October 2026Memory by quantisation and context
| Quant | 4,096 ctx | 8,192 ctx | 32,768 ctx |
|---|---|---|---|
| Q8_0 | 14.6 GB · fits 16 GB | 14.8 GB · fits 16 GB | 16.4 GB · fits 24 GB |
| Q5_K_M | 10.1 GB · fits 12 GB | 10.4 GB · fits 12 GB | 12 GB · fits 16 GB |
| Q4_K_M | 8.5 GB · fits 12 GB | 8.7 GB · fits 12 GB | 10.3 GB · fits 12 GB |
Gemma 4 12B is the newer model in this family. It needs about 8.2 GB at the same setting.
Weights at this quant: 7.1 GB at Q4. Every extra 1,000 tokens of context adds about 0.066 GB of KV cache at FP16, on top of a fixed 0.336 GB that the sliding-window layers hold once their window is full. Try other settings in the VRAM calculator.
About the model
Gemma 3 12B has 12.2B parameters. It has 48 layers with 8 KV heads of dimension 256; five in every six use sliding-window attention and only remember the last 1,024 tokens, so 8 global layers carry the long context. Google's 12B vision-language model from March 2025. It introduced the five-local-to-one-global layer pattern that Gemma 4 kept, which is why its long-context memory is modest for a 2025 model.
Text and images 131,072-token window Gemma Terms of Use March 2025
Reasoning. Answers directly; there is no reasoning mode.
A runtime without a sliding-window cache stores every layer at full length and needs far more than this at long context.
Native context window: 131,072 tokens. Licence: Gemma Terms of Use, a custom licence with its own conditions; read it before commercial use. Weights published in March 2025.
Good at
- Image understanding on a 12 GB card, with a mature ecosystem of quantisations and fine-tunes.
- Multilingual work: the family covers over 140 languages.
Watch out for
- Gemma Terms of Use, not Apache: a custom licence with a prohibited-use policy. Gemma 4 dropped it.
- Gemma 4 12B replaces it, adds audio and video, and needs less memory for the same context.
Similar models
The models that need about the same memory as Gemma 3 12B at Q4_K_M and 8k, which makes them the real alternatives on whatever card you have. The context column is what each extra 1,000 tokens costs, and it is where models of the same size differ most.
| Model | Size | Needs | Per 1k context | Window | Released |
|---|---|---|---|---|---|
| Gemma 3 12B | 12.2B | 8.7 GB | 0.066 GB | 128k | 2025-03 |
| Gemma 4 12B | 12B | 8.2 GB | 0.016 GB | 256k | 2026-05 |
| Ministral 3 14B | 14B | 10.3 GB | 0.164 GB | 256k | 2025-12 |
| Granite 4.2 8B | 8.8B | 7.2 GB | 0.164 GB | 128k | 2026-08 |
| Qwen3 14B | 14.8B | 10.8 GB | 0.164 GB | 40k | 2025-04 |
See every model compared, which models fit on 8 to 96 GB, or the cheapest hour of every cloud card.
Questions
- Can I run Gemma 3 12B on a 24 GB card?
- Yes, at Q4_K_M and 8k context it needs about 8.7 GB, which fits a 24 GB card with room.
- How much VRAM does Gemma 3 12B need at Q8?
- About 14.8 GB at 8k context, or 16.4 GB at 32k. Q8 is near-lossless; use it when it fits.
- What is the context length of Gemma 3 12B?
- 131,072 tokens natively. Holding all of it at Q4_K_M takes about 16.8 GB, of which 8.9 GB is context.
- Why does Gemma 3 12B need so little memory for long context?
- Because most of its layers do not keep a cache that grows. It has 48 layers with 8 KV heads of dimension 256; five in every six use sliding-window attention and only remember the last 1,024 tokens, so 8 global layers carry the long context. At 32k tokens the context costs about 2.5 GB; if all 48 layers kept a conventional full-length cache, the same conversation would cost about 12.9 GB. That saving depends on the runtime implementing the sliding-window cache. llama.cpp and vLLM do; a build that does not will use far more.
- Is there a newer model than Gemma 3 12B?
- Yes: Gemma 4 12B, published in May 2026. At the same Q4_K_M and 8k setting it needs about 8.2 GB against 8.7 GB for this one.
- How fast is Gemma 3 12B on an RTX 4090?
- At most about 100 tokens per second at Q4, single stream, which is faster than you can read. That is a ceiling worked out from the card's memory bandwidth, not a measurement.
- Architecture values from config.json in google/gemma-3-12b-it on Hugging Face (the repository is licence-gated: the card is public, the file opens once you accept the terms), checked 16 October 2026. Every figure on this page is computed from those values; check them before buying hardware for this model.
- Estimates assume a single conversation on a card that is otherwise free. A desktop on the same GPU takes 0.5 to 2 GB.